social contagion modeling

This study addresses the problem of algorithmic sycophancy induced by large language models, which can trap populations in spirals of erroneous collective beliefs. To counteract this cognitive bias, the authors construct a networked stochastic dynamical system incorporating a minority of topologically central “teacher” nodes designed to correct group-level misperceptions. By employing degree-weighted mean-field approximation, they reduce the high-dimensional Langevin equations to a macroscopic drift equation, thereby offering the first analytical framework that integrates statistical physics with social network theory to elucidate AI-induced sycophancy. Key contributions include an analytical solution for the critical tipping time based on saddle-node bifurcation, a proof that centralized rapid intervention outperforms distributed slow strategies, and demonstration of universal data collapse and theoretical bounds across diverse network topologies. Under strict budget constraints, the work further derives an optimal intervention policy.

This study challenges the conventional assumption that emotionally stable individuals dominate collective emotional contagion, instead investigating whether emotional instability—characterized by high neuroticism and low conscientiousness—enhances an individual’s capacity to drive group-level emotional diffusion. Method: Leveraging 733,534 daily affective entries from 38 professionals over 30.5 months, we applied Granger causality network reconstruction to conduct longitudinal analysis of emotion transmission dynamics. Results: Neuroticism (r = 0.478, p = 0.002) and low conscientiousness (r = −0.512, p = 0.001) significantly predicted individual emotional transmission strength. The basic emotional reproduction number (R₀) reached 15.58—far exceeding the epidemic threshold—while group-level affective variance increased by 22.9% over time. We introduce the “neuroticism paradox” and an affective epidemiology framework, providing the first empirical evidence that highly emotionally labile individuals function as “super-spreaders” of collective affect, thereby redefining the theoretical relationship between personality traits and affective leadership in groups.

This work reveals that infinitesimal structural perturbations—such as the addition or deletion of a single edge—in social networks can trigger global collective radicalization, highlighting the extreme fragility of social influence systems under generative AI–driven interventions. Method: We propose a robust analytical framework based on Dynamic Structure Functions (DSFs), the first application of DSF theory from control engineering to social dynamics, enabling quantitative identification of minimal-norm critical perturbations capable of destabilizing the system. Contribution/Results: Rigorous analysis on the Taylor social influence model demonstrates that arbitrarily small perturbations can violate linear stability, inducing unbounded growth in node sentiment states. We characterize a novel “imperceptible–lethal” vulnerability mechanism: perturbations undetectable at local scales yet sufficient to induce systemic collapse. This work provides a new theoretical foundation and quantitative criteria for assessing societal risks arising from algorithmic interventions in digital ecosystems.

Existing models of complex contagion struggle to capture the interplay among individual preferences, local social influence, and global sentiment, and offer limited insight into the critical thresholds governing phase transitions in viral spread. This work proposes a unified cascade model that embeds both ideas and network nodes into a shared high-dimensional feature space. Node state updates are driven by a decision function integrating transmission affinity, local reinforcement, and global activation, yielding an efficiently samplable Markovian cascade process. The model reveals, for the first time, how the dynamic balance between local and global influences critically determines cascade success or failure, and demonstrates that early-stage growth patterns can effectively predict phase transitions. Comprehensive experiments analyze cascade distributions, latent dynamics, parameter sensitivity, and critical behavior, establishing a new paradigm for studying complex contagion mechanisms.

Understanding how network topology and agent personality traits jointly shape the evolution of cooperation remains an open challenge. Method: We propose the first networked game simulation framework integrating MBTI-based personality modeling with large language model (LLM)-driven agents, enabling personality-informed behavioral simulation; we conduct repeated Prisoner’s Dilemma multi-agent simulations on small-world and scale-free networks. Contribution/Results: We demonstrate that cooperation evolution cannot be predicted solely from pairwise interactions but emerges from the coupling between personality spatial distribution and network structure. Specifically, cooperation levels significantly increase when prosocial personalities concentrate at hub nodes in scale-free networks, whereas small-world topologies consistently suppress cooperation. To support reproducible research, we open-source the NetworkGames simulation framework—a novel methodological foundation for investigating personality–structure coevolutionary effects in complex adaptive systems.

This work addresses the limitation of existing generative agents in social media simulation, which often fail to model users’ behavioral tendencies, leading to homogeneous interaction patterns. To overcome this, the study introduces behavioral traits as an explicit, independent representational layer within the generative agent framework. A parameterized mechanism modulates each agent’s propensity toward specific actions—such as posting, sharing, commenting, liking, or remaining inactive—emphasizing that “how to act” is as critical as “who one is.” Large-scale multi-agent simulations involving 980 agents powered by large language models demonstrate that this approach effectively preserves consistent yet heterogeneous participation patterns across individuals. The method successfully replicates key dynamics of content propagation observed on real-world social platforms, advancing social simulation from static user profiles toward dynamic behavioral evolution.

This study establishes a microfoundational link between cross-sectional network models and their underlying generative behavioral mechanisms. Building on a continuous-time stochastic choice framework, it proposes a general modeling approach that accommodates non-network actors and multilateral relational constraints, yielding an exponential-family representation under equilibrium conditions to facilitate individual preference estimation. The key innovation lies in the natural decomposition of graph potential into a preference component reflecting agent utilities and an entropy component encoding tie-formation rules, thereby unifying behavioral and statistical network modeling paradigms. The framework is successfully applied to analyze friendship networks in professional organizations and to model structural phase transitions in small groups, demonstrating its empirical validity and broad applicability.

This study investigates how stereotypes and affective polarization can spontaneously emerge in the absence of factual grounding or ideological conflict. To this end, we develop an agent-based dynamic belief network model that integrates social interaction and cognitive consistency mechanisms to simulate the evolution of individual beliefs and the formation of their associative structures. The model represents belief dynamics through causal and associative links and incorporates group identity as a driver of affective differentiation. Experimental results successfully reproduce the emergence of fact-free stereotypes and demonstrate how these subsequently fuel intergroup affective polarization. Our findings validate the pivotal role of purely social and cognitive mechanisms in bias formation and highlight the framework’s explanatory power and novelty in understanding irrational social fragmentation.

This study investigates whether large language models (LLMs) can faithfully simulate emotional diffusion patterns observed in real online communities—specifically Reddit. Method: We construct LLM-generated multi-turn dialogue graphs and systematically compare them against empirical Reddit social graphs across four dimensions: structural connectivity, interaction recurrence, affective dynamics, and community emergence. We integrate VADER and BERT-based sentiment analysis with graph-behavior joint modeling to quantify discrepancies. Contribution/Results: We provide the first empirical evidence that LLM-simulated graphs consistently exhibit linear, isolated chain structures and monotonic sentiment trajectories—lacking the high-density connectivity, sentiment reversals, and heterogeneous evolution characteristic of real networks. Consequently, LLM simulation induces significant reduction in affective diversity and severe class imbalance, degrading downstream graph prediction performance. This work reveals fundamental structural limitations of current LLMs in modeling socio-dynamic processes and establishes a critical evaluation benchmark and actionable direction for trustworthy, AI-driven social simulation.